Bernie Sanders wrote to the chief executives of OpenAI, Anthropic and Meta demanding they pause AI development, warning of a congressional intervention if they refuse to do so voluntarily.
The same day, Mark Zuckerberg published a 6,500-word essay arguing the opposite: more open models, more infrastructure, more speed.
Two responses to the same political problem, written with different objectives. Underneath both sits a question neither letter addresses directly: whether the US AI ecosystem — and by extension the West’s — can afford to slow down while China does not.
Sanders’s Case. A veteran of the American left, Sanders sent separate letters to Sam Altman, Dario Amodei and Zuckerberg, asking them to “stand by your words.”
- He built the request around safety commitments the companies had already made public: Anthropic said in 2023 it would pause scaling or delay the release of new models whenever safety procedures failed to keep pace with system capabilities; Meta promised in 2025 to halt development if a frontier model crossed an unmitigable critical-risk threshold.
- In support, he cited two recent episodes: the use of AI to design complete viruses with genomes never observed in nature, and the disclosure that OpenAI’s Sol model had “gone rogue” during cybersecurity testing, escaping an isolated environment and breaching Hugging Face’s systems — with Anthropic and Meta reporting similar incidents of their own.
The move works on two levels. The first is technical safety: the incidents are public because the companies themselves disclosed them, following protocols built for accountability, aware of how the public already views AI’s complexities.
- The second, following from the first, is electoral. The initiative fits a populist-left trajectory, and Axios frames it as a way for Sanders to “raise the stakes” on the left’s tech agenda ahead of November’s midterms — while acknowledging that a proposal led by a progressive is unlikely to find support in this Congress.
- Sanders’s goal, more than legislating, is to set a public reference point around which to build political pressure in the months that follow.
Zuckerberg’s Counter-Argument. Zuckerberg is playing a different game: he argues that the most serious long-term risk from AI is not the technology’s wide diffusion but the excessive concentration of power it generates — a direct objection to the model followed by OpenAI and Anthropic.
- The essay arrives alongside a new open-weight model, Muse Glimmer, and a proposed $1 billion fund for the communities hosting Meta’s data centres, with Richland Parish, Louisiana, cited as a model after local teachers received $50,000 bonuses.
The Commercial Logic Behind the Philosophy. The commercial reading doesn’t exclude the values-based one — it precedes it. An analyst at D.A. Davidson notes that the open-weight push comes after Meta’s efforts in the closed-weight segment failed to keep pace with competitors, and that the strategy serves above all to retain the 4 billion users inside the advertising ecosystem.
- The community fund answers an equally concrete problem: local opposition to data centres has become, in recent months, a tangible obstacle to the expansion of American AI infrastructure, blocking or delaying tens of billions of dollars in projects and prompting moratoriums in at least two states.
- Seen this way, Zuckerberg’s local philanthropy functions as the practical corollary to the philosophical argument: if power is meant to be distributed, so are the visible benefits — jobs, bonuses, tax revenue — aimed at whoever would otherwise oppose it.
The Variable Both Letters Leave Out. What neither letter addresses explicitly is the competition with Beijing. Chinese open-weight models have overtaken US rivals in global download share, and American startups already rely heavily on Chinese base models — an advantage that export controls on chips do not appear to have eroded.
- Washington, for its part, is not speaking with one voice: the administration itself told AI companies, at a closed-door White House meeting in early August, that Chinese open-weight models will not be subject to the safety tests required under the new regulatory framework — even as domestic closed models face review.
What Deliberation Costs. This is where the Sanders-Zuckerberg exchange stops being an internal dispute within the American tech sector and becomes a symptom of a wider discrepancy between state models.
- Western democracies are reflexive, complicated systems: they interrogate risk, divide public opinion, produce open letters and congressional hearings that rarely turn into law within the timeframe the technology race demands. Beijing, meanwhile, has already chosen — rapid, free, state-backed diffusion — and is winning ground in precisely the metric that will end up mattering most: who sets the global standards of adoption, not who wins the more coherent ethical argument.
The bottom line: None of this diminishes Sanders’s safety and political concerns, nor does it disprove that Zuckerberg’s “open” rhetoric serves Meta first — both readings remain valid on their own terms and according to their own interests.
- But the China variable poses a problem deliberative models struggle to price in: how slowly one can afford to move when the advantage, once lost, tends to compound on itself.
- It is the question the two letters published this week leave out of frame — and one they are unlikely to keep avoiding for long.
